基于三种空间预测方法的安庆市耕地土壤速效钾空间分布预测

以安徽省安庆市为研究区,选取环境变量因子(空间位置变量因子、地形变量因子、土壤变量因子、气候变量因子)作为变量因素,通过构建随机森林(Random Forest,RF)模型对研究区耕地土壤速效钾含量进行预测,并与普通克里金(Ordinary Kriging,OK)和反距离权重(Inverse Distance Weig...

Prediction of spatial distribution of available potassium in cultivated soil of Anqing city based on three spatial prediction methods

Taking Anqing City of Anhui Province as the research area,a Random Forest model(RF)is constructed by selecting environmental variable factors(space position variable factor,topographic variable factor,soil variable factor,climate variable factor)as the variable factor.The content of available potassium is predicted and compared with the results by traditional spatial prediction methods of Ordinary Kriging(OK)and Inverse Distance Weight(IDW).The results show that the prediction accuracy of the three methods for the spatial distribution of available potassium in the study area is RF>OK>IDW,where the mean absolute error(MAE)and root mean square error(RMSE)of the RF model and the coefficient of determination(R2)are 30.93 mg·kg^-1,41.31 mg·kg^-1 and 0.58,which are 3.36%,5.95%,6.71%,11.86%and 18.37%,23.40%higher than those of OK and IDW.The overall trends of the three spatial distribution prediction methods are consistent,showing a high distribution in the southeast and a low distribution in the northwest.In summary,the RF model can better predict the available potassium content of cultivated soil in Anqing.Latitude,annual average temperature,soil parent material,elevation,longitude,and annual average precipitation are the main factors that affect the accuracy of the RF model.